Please explain stable and unstable learning algorithms with examples and then categorize different classifiers into them. (For example NN, SVM, Linear classifier, FLD, KNN and decision tree are stable or unstable).
What do you mean by stable? Deterministic? Or robust against noise?
go to "http://en.wikipedia.org/wiki/Stability_(learning_theory)".
A stable algorithm is an algorithm that by changing inputs a little, class labels don't change.
Thanks for the link! But isn't the answer to your question already on there?
If it was there I don't ask this question.:)
I need to understand that all classifying methods are stable or unstable.
According to that Wikipedia page stability is the ability to generalize a problem. SVM is known to be designed for this kind of problem. Other techniques can be used as well without problems, but you always have to take into account that there are no guarantees for noisy inputs unless you train them to deal with noise. If you train them in noisy input data (populate your training set by perturbations) than it will learn to handle it. If stability assumes slight change in the observed world than no method can certainly deal with that without a proper re-training session (obviously, if the world is changing we need to adapt ourselves to the new discoveries).
I want to say one thing those things depend on the problem domain itself, where ur appliying all those techniques.
If learning algorithm is stable, changing input slightly won't put effect on the hypothesis. In unstable, it would.
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification